Skip to content

GitLab

Commit f32abe97 authored by Davis King's avatar Davis King
Browse files

Slightly changed solver code to avoid a wasted initial iteration.

parent 98d9703d
Hide whitespace changes
Inline Side-by-side
Showing with 2 additions and 2 deletions
dlib/optimization/optimization_solve_qp_using_smo.h
View file @ f32abe97
...@@ -511,7 +511,7 @@ namespace dlib ...@@ -511,7 +511,7 @@ namespace dlib
// We need to get an upper bound on the Lipschitz constant for this QP. Since that // We need to get an upper bound on the Lipschitz constant for this QP. Since that
// is just the max eigenvalue of Q we can do it using Gershgorin disks. // is just the max eigenvalue of Q we can do it using Gershgorin disks.
const T lipschitz_bound = max(diag(Q) + (sum_cols(abs(Q)) - abs(diag(Q)))); const T lipschitz_bound = max(diag(Q) + (sum_cols(abs(Q)) - abs(diag(Q))));
double lambda = 0; double lambda = 1;
unsigned long iter; unsigned long iter;
for (iter = 0; iter < max_iter; ++iter) for (iter = 0; iter < max_iter; ++iter)
{ {
...@@ -829,7 +829,7 @@ namespace dlib ...@@ -829,7 +829,7 @@ namespace dlib
std::vector<matrix<T,NR,NC,MM,L>> v(alphas), v_old(alphas.size()); std::vector<matrix<T,NR,NC,MM,L>> v(alphas), v_old(alphas.size());
double lambda = 0; double lambda = 1;
unsigned long iter; unsigned long iter;
// Now do the main iteration block of this solver. The coordinate descent method // Now do the main iteration block of this solver. The coordinate descent method
// we used above can improve the objective rapidly in the beginning. However, // we used above can improve the objective rapidly in the beginning. However,
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment